Claude Config

This repository holds my personal configuration for agentic coding. Unless you are a scientist or Julia user, it may not be relevant for you.

I'm making it public for the benefit of people in my laboratory and any other users who might want to learn from it. However, it is not intended to be a common resource, and issues/pull requests may be ignored. However, you are welcome to copy or fork this as you see fit (see LICENSE.md).

For me, this repo is my ~/.claude folder with a lot of required material git-ignored; if you want to use tools from here, either copy the useful bits into your own ~/.claude or rename your existing ~/.claude to a temporary name, clone your fork of this repo as your ~/.claude, and then manually copy all the missing pieces from your old renamed folder.

Key components

- CLAUDE.md: general instructions applicable in every session. Includes stance, programming language and style choices, MCP-agent debugging tips, and devops tips (commits, GitHub etiquette). If you use this, customize it extensively to your own preferences.

- rules/: focused guidance files that the harness auto-loads into every session, exactly like CLAUDE.md (they must live in ~/.claude/rules/ for this to happen — no @import is needed, but the behavior is a harness convention). Splitting these out of CLAUDE.md keeps each topic individually addressable, so a skill can cite it by path and other rules can cross-reference it with [[name]] links. Current rules cover code-comment discipline and Julia-specific correctness (generic indexing, @inbounds).

- hooks/: scripts invoked by the harness around tool calls. For example, no-plan-comments.py runs after every edit and reminds the agent if planning/history scaffolding has leaked into code comments. Hooks are *registered* in…